MSc Artificial Intelligence · University of Southampton
Building at the intersection of Knowledge Graphs, Retrieval-Augmented Generation, and LLM reasoning systems.
- MSc Dissertation — Investigating GNN topology encoding to improve multi-hop reasoning in Knowledge Graph RAG systems, extending the Reasoning on Graphs (RoG) framework. Evaluated on WebQSP and CWQ benchmarks.
- DPR-RAG Baseline — Implementing and validating dense passage retrieval baselines against published results as a controlled experimental foundation.
Previously built a production Hybrid RAG system integrating Neo4j knowledge graphs with community-based retrieval and prompt optimisation at Kovai.co. Also built an LLM-powered scientific literature analysis pipeline at Nutrify Today.
| Project | Description | Stack |
|---|---|---|
| Wikidata KG Query Engine | SPARQL-based knowledge graph querying — multi-hop traversal, aggregation, negation patterns | SPARQL, Python, Wikidata |
| Human Activity Recognition | End-to-end ML pipeline: K-Means → Random Forest → 1D CNN (93.3% accuracy, 5M+ samples) | Python, PyTorch, scikit-learn |
| Credit Card Fraud Detection | Leakage-safe pipeline for rare-event classification (0.17% fraud rate) — 6 models compared | Python, scikit-learn, XGBoost, LightGBM |
| Hybrid RAG + Knowledge Graph | RAG system combining graph-structured retrieval with vector search and LLM inference | Python, Neo4j, LangChain, FAISS |
AI/ML RAG · Knowledge Graphs · LLM Reasoning · Prompt Engineering
PyTorch · HuggingFace · scikit-learn · FAISS · LangChain
Languages Python · SPARQL · Cypher (Neo4j)
Databases Neo4j · MongoDB · FAISS
Tools Git · Jupyter · VS Code · Streamlit · Ollama
- MSc Artificial Intelligence — University of Southampton (2025–2026)
- B.Tech, AI & Data Science — SKCET, India (2021–2025)
harshng7@gmail.com · LinkedIn · Southampton, UK